ATTENTION: pydalle is now no longer supported as there is an official API. With the recent addition of a captcha to the OpenAI login page, the automated approach to get a token used in this library won't work any longer anyway. Thanks for using pydalle while it was relevant!

pydalle: A DALL·E 2 API Wrapper for Python

This library provides basic programmatic access to the DALL·E 2 API.

The intent of this library is to provide researchers with a means to easily layout results from DALL·E 2 into a jupyter notebook or similar.

pydalle has two main modes of use:

• pydalle.Dalle: This is the main class of the library. It provides a user-friendly interface to the DALL·E 2 API. Read more here.
• pydalle.imperative.api.labs: This module provides a set of lower-level functions that can be used to interact with the DALL·E 2 API. Read more here.

Installation

Install with all dependencies

pip install pydalle[all]     # Install all dependencies, recommended for most users

Pick and choose your dependencies

pip install pydalle          # Just install the library with no optional dependencies
pip install pydalle[sync]    # Also installs requests (for synchronous networking)
pip install pydalle[async]   # Also installs aiohttp and aiofiles  (required for async networking / file handling)
pip install pydalle[images]  # Also installs Pillow and numpy (required for help with image processing)

Tips

• Get access by signing up for the DALL·E 2 waitlist.

• Ensure your usage of DALL·E 2 abides by DALL·E 2's content policy and terms of use.

• Be mindful about how easy this library makes it for you to spend your money / DALL·E 2 credits.

Getting Started

Once you have installed pydalle, you can start using it by importing it and creating a Dalle object. You can find all the available methods on the Dalle class.

import os

from pydalle import Dalle

OPENAI_USERNAME = os.environ.get('OPENAI_USERNAME')
OPENAI_PASSWORD = os.environ.get('OPENAI_PASSWORD')


def main():
    client = Dalle(OPENAI_USERNAME, OPENAI_PASSWORD)
    print(f"Client created. {client.get_credit_summary().aggregate_credits} credits remaining...")
    tasks = client.get_tasks(limit=5)
    print(f"{len(tasks)} tasks found...")

    print("Attempting to download a generation of the first task and show off some built-in helpers...")
    if tasks and tasks[0].generations:
        example = tasks[0].generations[0].download()
        example.to_pil().show()  # Convert the image to a PIL image and show it
        example.to_pil_masked(x1=0.5, y1=0, x2=1, y2=1).show()  # Show a version with left side transparent (for edits)
        example.to_pil_padded(0.5).show()  # Show w/ 50% padding around the image, centered at (50%, 50%)
        example.to_pil_padded(0.4, cx=0.25, cy=0.25).show()  # Show w/ 40% padding, centered at (25%, 25%)

    print("Attempting to do a text2im task...")
    completed_text2im_task = client.text2im("A cute cat")
    for image in completed_text2im_task.download():
        image.to_pil().show()

    print("Attempting to create variations task on the first cat...")
    first_generation = completed_text2im_task.generations[0]
    completed_variation_task = first_generation.variations()
    first_variation = completed_variation_task.generations[0]
    first_image = first_variation.download()
    first_image.to_pil().show()

    print("Attempting to create inpainting task and showing the mask...")
    # Make the right-side of the image transparent
    mask = first_image.to_pil_masked(x1=0.5, y1=0, x2=1, y2=1)
    mask.show("inpainting mask")
    completed_inpainting_task = first_generation.inpainting("A cute cat, with a dark side", mask)
    for image in completed_inpainting_task.download():
        image.to_pil().show()


if __name__ == '__main__':
    main()

For an equivalent async code example, see examples/dev_client_async.py.

For examples of the low-level API and using this in a notebook, see the examples/low_level directory.